Noam Shazeer Leaves Google for OpenAI: What It Means for Gemini and ChatGPT (June 18, 2026)

Noam Shazeer — Transformer co-inventor, Google DeepMind vice president of engineering, and Gemini co-lead — confirmed on June 17, 2026 that he is leaving Google to join OpenAI ahead of its planned IPO. Less than two years after Google paid roughly $2.7 billion to bring him back from Character.AI, the departure is the highest-profile Google-to-OpenAI move of the AI era so far. Here’s what we know and what it means for Gemini, ChatGPT, and the Google–OpenAI talent war.

Last verified: June 18, 2026.

TL;DR

• Confirmed: Reuters, CNBC, and US News reported on June 17, 2026 that Shazeer is moving from Google to OpenAI.
• Context: Google reportedly paid ~$2.7B in 2024 to bring him back from Character.AI; he’s been Gemini co-lead since.
• Pedigree: Shazeer co-authored the 2017 Transformer paper that underpins every modern LLM.
• Timing: OpenAI filed confidentially for IPO June 8, 2026. Anthropic filed June 1. The pre-IPO talent war is on.
• Impact on Gemini 3.5 Pro: Negligible short term. Long term, the Gemini 4 cycle has to absorb the loss.
• Impact on OpenAI: Strong pre-IPO research-bench signal; expect Shazeer in the S-1 narrative.

What was announced on June 17, 2026

Noam Shazeer publicly confirmed his departure from Google to OpenAI on Wednesday, June 17, 2026. The news was reported simultaneously by Reuters (“Google Gemini co-lead Noam Shazeer to join IPO-bound OpenAI”), CNBC (“Google Gemini co-lead Noam Shazeer leaves for OpenAI”), and US News & World Report. As of the morning of June 18, 2026, neither Google nor OpenAI has published an official blog post detailing the move, but multiple outlets independently confirmed it.

Shazeer was at Google as:

• Vice president of engineering, Google DeepMind
• Co-lead of the Gemini model line

He is departing to join OpenAI. Neither company has publicly named his specific project.

Why Shazeer matters

Shazeer is not a normal senior engineer. He is one of the eight co-authors of the 2017 paper “Attention Is All You Need,” which introduced the Transformer architecture that underpins every major modern large language model: GPT, Claude, Gemini, Llama, Mistral, DeepSeek, Kimi K2.7 Code (released June 12, 2026), and the entire downstream ecosystem of AI agents and coding tools.

He also has a track record of building scaling-era systems:

Year	Role	Significance
2000–2021	Google (various roles)	Co-author of the Transformer (2017); core contributor to early TPU and large-model infra.
2021–2024	Co-founder & CEO, Character.AI	Built consumer-facing conversational models at scale.
2024	Returned to Google via ~$2.7B Character.AI deal	Became VP of engineering at Google DeepMind.
2024–2026	Co-lead, Gemini	Co-led Gemini 2.5 Pro, Gemini 3 Pro, Gemini 3.5 Pro generation.
June 17, 2026	Announces move to OpenAI	Joins IPO-bound OpenAI.

The dollar-cost context is what makes this story unusually loud. Google reportedly paid roughly $2.7 billion in 2024 to acquire Character.AI’s IP and bring Shazeer back. Roughly 22 months later, he is going to the company Google most directly competes with. That math is going to be cited in retention discussions across the frontier labs for the rest of 2026.

What this means for Gemini

Short term, almost nothing visible. Gemini 3.5 Pro hit general availability in early June 2026 and is shipping inside Google products today. Demis Hassabis remains CEO of Google DeepMind. Jeff Dean, Oriol Vinyals, and the rest of the Gemini leadership are in place. The product roadmap from Google I/O 2026 — Create My Widget, Magic Pointer, Android XR Gemini features, Googlebooks running Aluminium OS — is unaffected.

Long term, the Gemini 4 cycle has to absorb the loss. Gemini 4 is currently targeted for late 2026, and a model co-lead departing mid-cycle is the kind of disruption that typically forces other internal leads to absorb scope or accelerate sub-projects. Google has navigated this before — Shazeer left in 2021, Gemini still shipped, and the company recruited him back. The honest read for Gemini 4 is not “will it ship” but “will its architectural ambition have to be scoped down in his absence.” Watch for any quiet narrowing of the Gemini 4 feature set in Google’s autumn 2026 communications.

What this means for OpenAI

OpenAI filed confidentially for a US IPO on June 8, 2026 — a week after Anthropic filed on June 1, 2026. The S-1 process is now actively underway. Adding a Transformer co-inventor and an active Gemini co-lead to the technical team during the run-up to that listing strengthens OpenAI’s pre-IPO research-bench narrative in three ways:

1. S-1 talent disclosure. Senior research hires of this profile typically appear in the management-and-key-personnel section of the S-1 and in roadshow talking points.
2. Self-improvement narrative. OpenAI has been pushing the line, since GPT-5.3-Codex, that its frontier systems are increasingly self-improving. Adding architectural muscle reinforces that story for institutional investors.
3. Recruitment flywheel. A move of this size publicly tilts the talent gradient back toward OpenAI for the next 12 months. Expect competitive offers from Anthropic, Meta, and xAI to rise in response.

What this means for the Google–OpenAI talent war

This is the highest-profile single move from Google to OpenAI so far. Prior cross-flows between the two labs typically involved researchers below the VP-of-engineering tier, or moves into Anthropic rather than directly to OpenAI. Shazeer’s seniority, his Transformer pedigree, and the $2.7B retention package make this move materially different. Expect three downstream effects in the next 6–12 months:

• Comp inflation at the senior level. Frontier-lab packages for VP-of-engineering and director-of-research roles have probably just gone up.
• Equity-and-IPO-driven recruitment. OpenAI’s IPO equity is now a real recruiting weapon. Anthropic’s October 2026 IPO target is in the same position.
• Google’s defensive recruiting. Watch for Google to make at least one offsetting senior hire in the next 60 days to keep the Gemini 4 narrative intact.

How to think about this if you build with these models

• If you’re building on Gemini 3.5 Pro: No change is needed. The model line is stable, the team is large, and Gemini 4 is still on the roadmap. Don’t make a switching decision off a single talent move.
• If you’re building on GPT-5.5 or Codex: The OpenAI research line is reinforced, not weakened. Continue as planned.
• If you’re a multi-provider shop: The pragmatic posture remains the same — keep your stack model-portable (OpenAI Agents SDK, Claude Agent SDK, Vertex AI), watch the GPT-6 and Gemini 4 launch windows, and re-evaluate when those models actually ship.

Sources and verification

• Reuters, “Google Gemini co-lead Noam Shazeer to join IPO-bound OpenAI,” June 17, 2026.
• CNBC, “Google Gemini co-lead Noam Shazeer leaves for OpenAI,” June 17, 2026.
• US News & World Report, “Google Gemini Co-Lead Noam Shazeer to Join IPO-Bound OpenAI,” June 17, 2026.
• OpenAI IPO confidential filing: Reuters, June 8, 2026.
• Anthropic IPO confidential filing: New York Times, June 1, 2026.
• Character.AI / Google deal valuation (~$2.7B): Storyboard18 reporting, June 17, 2026.

Related pages on andrew.ooo

• OpenAI September IPO vs Anthropic October Race — the two listings in detail.
• Anthropic IPO October 2026 Timeline — Anthropic’s S-1 path.
• Gemini 3.5 Pro GA vs Claude Fable 5 vs GPT-5.6 Late June 2026 — current model landscape.
• Android 17 Gemini Omni vs iOS 26 Siri AI — Gemini in consumer products.

This page summarises a fast-moving June 17–18, 2026 news event. We will update if OpenAI or Google publish formal statements.